def learn(X, Y):
pca = None
dictionary = None
model = None
# Data augmentation
if DO_DATA_AUGMENTATION:
print("Augmenting data")
X, Y = transform_T(X, Y)
print("Number of samples augmented to {}".format(X.shape[0]))
# Dictionary learning
if DO_DICTIONARY_LEARNING:
dictionary = Dictionary(n_atoms=128, atom_width=16)
if dictionary.weights_available:
print("Loading dictionary")
dictionary.load()
else:
print("Learning dictionary")
tic = time.time()
dictionary.fit(X)
dictionary.save()
print("Dictionary learned in {0:.1f}s".format(time.time() - tic))
print("Getting dictionary representation")
X = dictionary.get_representation(X)
# PCA
if DO_PCA:
tic = time.time()
print("Applying PCA")
n_components = 100
pca = PCA(n_components=n_components)
X = pca.fit(X, scale=False)
print("Variance explained: {:.2f}".format(np.sum(pca.e_values_ratio_)))
print("PCA applied in {0:.1f}s".format(time.time() - tic))
# Training
print("Starting training")
tic = time.time()
model = KernelSVM(C=1, kernel='rbf')
model.train(X, Y)
print("Model trained in {0:.1f}s".format(time.time() - tic))
return pca, dictionary, model
def create(data_manager, components_limit=0):
'''
Creates statistical model by aligning models and performing PCA
:param data_manager: data manager providing training data
:param components_limit: Limit how much PCA components should be found (0 == all)
:return: resulting PCA
'''
teeth = data_manager.get_all_teeth(True)
mean_shape = deepcopy(teeth[0])
assert isinstance(mean_shape, Tooth)
mean_shape.move_to_origin()
mean_shape.normalize_shape()
error = float("inf")
while error > 0.05:
meanAcum = np.zeros(mean_shape.landmarks.shape)
for i in range(0, len(teeth)):
teeth[i].align(mean_shape)
meanAcum += teeth[i].landmarks
new_mean_shape = Tooth(meanAcum / len(teeth))
new_mean_shape.align(mean_shape)
error = new_mean_shape.sum_of_squared_distances(mean_shape)
mean_shape = new_mean_shape
# Realign all teeth with final mean shape
for i in range(0, len(teeth)):
teeth[i].align(mean_shape)
data = np.zeros((len(teeth), teeth[0].landmarks.size))
for i, tooth in enumerate(teeth):
data[i, :] = tooth.landmarks.flatten()
pca = PCA()
pca.train(deepcopy(data), components_limit)
return pca
def phase1(self, data, trueLabels):
step1ResultsFolder = Path(self.config["resultsDir"]) / "phase1"
step1ResultsFolder.mkdir(exist_ok=True, parents=True)
# step 2 plot the original dataset (picked N_COMPONENTS dims with highest variance)
stds = data.std(axis=0)
dims = np.argsort(stds)[::-1][:N_COMPONENTS]
Visualizer.labeledScatter3D(data[:, dims], trueLabels, path=step1ResultsFolder / f"{N_COMPONENTS}_dims_originalScatter.png")
# step 3
dataMean = np.mean(data, axis=0)
print(f"Original data mean: {dataMean}")
# step 4, 5, 6, 7
pca = PCA(n_components=N_COMPONENTS, print_=True)
reducedData = pca.fit_transform(data)
# step 8
Visualizer.labeledScatter3D(reducedData, trueLabels, path=step1ResultsFolder / f"{N_COMPONENTS}_dims_pcaScatter.png")
# step 9
reconstructedData = pca.inverse_transform(reducedData)
Visualizer.labeledScatter3D(reconstructedData[:, dims], trueLabels, path=step1ResultsFolder / f"{N_COMPONENTS}_dims_reconstructedScatter.png")
return reducedData
from src.pca import PCA
# custom One-Versus-Rest SVM
from src.ovr import OVR
SHAPE = (46, 56)
M = 121
standard = False
data = fetch_data(ratio=0.8)
X_train, y_train = data['train']
D, N = X_train.shape
pca = PCA(n_comps=M, standard=standard)
W_train = pca.fit(X_train)
X_test, y_test = data['test']
I, K = X_test.shape
W_test = pca.transform(X_test)
params = {'C': 1, 'gamma': 2e-4, 'kernel': 'linear'}
ovr = OVR(**params)
ovr.fit(W_train, y_train)
y_hat = ovr.predict(W_test[::-1]).ravel()
# return features and labels
return unsupervisedFeatures, y
def parseArguments():
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--jsonConfig", type=Path, required=True)
return parser.parse_args()
if __name__ == "__main__":
args = parseArguments()
with open(args.jsonConfig, 'r') as f:
config = json.load(f)
step1ResultsFolder = Path(config["resultsDir"]) / "phase2"
step1ResultsFolder.mkdir(exist_ok=True, parents=True)
data, trueLabels = loadArffFile(Path(config["path"]))
data = data.to_numpy()
pca = PCA(data.shape[1])
reducedData = pca.fit_transform(data)
print(len(pca.varianceRatios))
print(f"Explained Ratio Variance {np.round(pca.varianceRatios, 2)}")
#print(f"Explained Ratio Variance {np.round(pca.explained_variance_ratio_, 2)}")
#print(f"Sum {np.sum(pca.explained_variance_ratio_)}")
#print(f"Mean {pca.mean_}")
#print(f"Variance {ipca.var_}")
#0print(f"Noise Variance {pca.noise_variance_}")
#Visualizer.labeledScatter3D(reducedData, trueLabels, path=step1ResultsFolder / f"caca2_dims_pcaScatter.png")
# prettify plots
plt.rcParams['figure.figsize'] = [12.0, 9.0]
sns.set_palette(sns.color_palette("muted"))
_palette = sns.color_palette("muted")
sns.set_style("ticks")
M = 121
standard = True
data = fetch_data(ratio=0.8)
X_train, y_train = data['train']
D, N = X_train.shape
pca = PCA(n_comps=M, standard=standard)
W_train = pca.fit(X_train)
X_test, y_test = data['test']
I, K = X_test.shape
W_test = pca.transform(X_test)
params = {'gamma': 2e-4, 'kernel': 'linear'}
fine = 5
# validate OVR
C_ovr = np.logspace(-5, 10, fine)
accuracy_ovr = []